VLSI and Neural Net Systems (VNNS) Laboratory

VLSI Auditory Models

Analog Circuits for Auditory Sound Source Localization using Current Mode Techniques, Christian Karl, 2006.

This dissertation focuses on the analog VLSI implementation of auditory nerve models using current mode circuit design techniques. The target application of these chips is sound source localization, a task difficult to accomplish using standard digital signaling processing methods, especially in a reverberant environment. The models and the resulting circuitry have not been previously used commercially. In general, the usage of such models requires a dramatic rethinking of how to process sound signals. The approach itself is called "biomimetic"; that is to say, mimicking nature's biological systems, in this case the mammalian auditory system. The usage of a biomimetic scheme for sound processing is new and little or no circuitry has been developed to date to support such processing. This dissertation shows how appropriate circuits can be built, highlighting the importance of accuracy using analog VLSI.

A 64 analog channel system design comprised of 512 very low frequency current mode integrators and 128 current mode multipliers is presented. The design is constrained by the available power budget and layout area. Under these constraints a system is presented that is suitable for sound source localization with sensor spacing of eight inches and above. The main building blocks of a current mode auditory channel have been designed, fabricated and tested. The measured results from the test chips agree with the theoretical predictions as well as with the results from circuit and Monte Carlo simulations. The dissertation highlights the importance of matching constraints in current mode circuitry when designing massively-parallel, non-linear structures. Using Monte Carlo methods as well as a Design Of Experiment (DOE) approach, an analysis has been done, which reveals how to limit mismatch-induced timing jitter. The non-linear circuit behavior was characterized and methods have been elaborated, which will theoretically produce near 100% design yield under reasonable process corner limitations.

VLSI Auditory Models

VLSI Auditory Models

Last Modified: Sunday October 21, 2007